Comprehensive cross-platform comparison of methods for non-invasive EGFR mutation testing: results of the RING observational trial Atocha Romero1,2 , Eloisa Jantus-Lewintre3,4,5, Beatriz Garcıa-Pela ez6, Ana Royuela7, Amelia 8 9 10 rez Altozano , Oscar Juan Vidal12, Pilar Diz13, Manuel Insa , Patricia Cruz , Ana Collazo , Javier Pe 11 14 Cobo , Berta Herna ndez , Sergio Va 15 zquez Estevez16, Gretel Benıtez17, Maria Guirado18, Margarita 19 Majem , Reyes Bernabe  , Ana Laura Ortega21, Ana Blasco3, Joaquim Bosch-Barrera22, Jose M. 20 Jurado , Jorge Garcıa Gonza 23 lez24 , Santiago Viteri25, Carlos Garcia Giron26, Bartomeu Massutı , Ana Lopez Martın , Alejandro Rodriguez-Festa1, Silvia Calabuig-Farin 27 28 ~ as3,4,29, Miguel  Angel Molina-Vila 6 and Mariano Provencio 1,2 1 Liquid Biopsy Laboratory, Biomedical Sciences Research Institute Puerta de Hierro-Majadahonda, Madrid, Spain 2 Medical Oncology Department, Hospital Universitario Puerta de Hierro-Majadahonda, Madrid, Spain 3 CIBERONC, Madrid, Spain 4 Mixed Unit TRIAL, Prıncipe Felipe Research Center & General University Hospital of Valencia Research Foundation, Spain 5 Biotechnology Department, Universitat Polite cnica de Valencia, Spain 6 Laboratory of Oncology/Pangaea Oncology, Quiro n-Dexeus University Hospital, Barcelona, Spain 7 Biostatistics Unit, CIBERESP, Hospital Universitario Puerta de Hierro-Majadahonda, Madrid, Spain 8 Hospital Clınico Universitario de Valencia, Spain 9 Hospital la Paz, Madrid, Spain 10 Hospital Universitario Sanchinarro, Madrid, Spain 11 Hospital Virgen de los Lirios, Valencia, Spain 12 Hospital Universitario La Fe, Valencia, Spain 13 Complejo Asistencial Universitario de Leo n, Spain 14 Hospital Regional Universitario, Malaga, Spain 15 Complejo Hospitalario de Navarra, Spain 16 Hospital Universitario Lucus Augusti, Lugo, Spain 17 Complejo Hospitalario Universitario Insular de Gran Canaria, Las Palmas, Spain 18 Hospital General Universitario de Elche, Alicante, Spain 19 Hospital de la Santa Creu i Sant Pau, Barcelona, Spain 20 Hospital Virgen del Rocıo, Sevilla, Spain 21 Complejo Hospitalario de Jae n, Spain 22 Hospital Dr. Josep Trueta-ICO Girona, Spain 23 Hospital Universitario Clınico San Cecilio, Granada, Spain 24 Hospital Clınico Universitario de Santiago, A Corun ~a, Spain 25 Instituto Oncolo gico Dr. Rosell, Hospital Universitario Dexeus, Grupo Quiron Salud, Barcelona, Spain 26 Hospital Universitario de Burgos, Spain 27 Hospital General Universitario Alicante, Spain 28 Hospital Severo Ochoa, Legane s, Madrid, Spain 29 Department of Pathology, Universitat de Vale ncia, Spain Keywords Several platforms for noninvasive EGFR testing are currently used in the circulating free DNA; epidermal growth clinical setting with sensitivities ranging from 30% to 100%. Prospective factor receptor; NGS; non-small-cell lung studies evaluating agreement and sources for discordant results remain cancer; osimertinib; tyrosine kinase inhibitor lacking. Herein, seven methodologies including two next-generation Abbreviations BEAMing, beads, emulsion, amplification, and magnetics; cfDNA, circulating free DNA, cell-free DNA; cobas, cobas EGFR Mutation Test v2 (Roche Diagnostics); ctDNA, circulating tumor DNA; CUSUM, cumulative sum; ddPCR, droplet digital polymerase chain reaction; dPCR, digital polymerase chain reaction; EGFR, epidermal growth factor receptor; FFPE, formalin-fixed, paraffin-embedded; ICC, intraclass correlation coefficient; MAF, mutant allele frequency; NGS platforms, Ion S5™ XL and GeneRead™; NGS, next-generation sequencing; NSCLC, non-small-cell lung cancer; PNA-Q-PCR, peptic nucleic acid probe-based real-time polymerase chain reaction; Therascreen, Therascreen EGFR Plasma RGQ PCR Kit (QIAgen); TKI, tyrosine kinase inhibitor. Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 43 This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Liquid biopsy-based cross-platform comparison A. Romero et al. Correspondence sequencing (NGS)-based methods, three high-sensitivity PCR-based plat- A. Romero, E. Jantus-Lewintre and M. A.  forms, and two FDA-approved methods were compared using 72 plasma Molina-Vila, Laboratorio Biopsia Lıquida, samples, from EGFR-mutant non-small-cell lung cancer (NSCLC) patients Oncologıa Me dica, Hospital Universitario progressing on a first-line tyrosine kinase inhibitor (TKI). NGS platforms Puerta de Hierro, C/Manuel de Falla 1, 28222 Majadahonda, Madrid, Spain as well as high-sensitivity PCR-based methodologies showed excellent Tel: +34 91 1917769 agreement for EGFR-sensitizing mutations (K = 0.80–0.89) and substantial E-mails: atocha10@hotmail.com (AR); agreement for T790M testing (K = 0.77 and 0.68, respectively). Mutant ejantus@btc.upv.es (EJ-L); and allele frequencies (MAFs) obtained by different quantitative methods mamolina@panoncology.com (MAM-V) showed an excellent reproducibility (intraclass correlation coefficients 0.86– 0.98). Among other technical factors, discordant calls mostly occurred at Trial registration: NCT03363139 mutant allele frequencies (MAFs) ≤ 0.5%. Agreement significantly (Received 28 June 2020, revised 1 October improved when discarding samples with MAF ≤ 0.5%. EGFR mutations 2020, accepted 22 October 2020, available were detected at significantly lower MAFs in patients with brain metas- online 13 November 2020) tases, suggesting that these patients risk for a false-positive result. Our results support the use of liquid biopsies for noninvasive EGFR testing and doi:10.1002/1878-0261.12832 highlight the need to systematically report MAFs. [Correction added on 21 December 2020, after first online publication: Peer review history is not available for this article, so the peer review history statement has been removed.] to the sensitivity of the method used, which may limit 1. Introduction the access to targeted therapies. Tyrosine kinase inhibitors (TKIs) have dramatically Currently, several platforms are available for noninva- changed the outcome of patients with EGFR-positive sive EGFR testing in blood, some of which have received non-small-cell lung cancer (NSCLC) [1–4]. Osimer- approval from the U.S. Food and Drug Administration tinib, a third-generation TKI, is currently the standard (FDA) as well as European Medicines Agency (EMA) as of care as second-line treatment in patients with companion kits for EGFR TKIs. Notably, the reported T790M-positive tumors at progression to first- or sec- sensitivities of the different assays for EGFR mutation ond-generation EGFR TKI [5] as well as for first-line detection in the cfDNA from patients with advanced treatment of EGFR-positive NSCLC patients [6]. Bio- NSCLC vary as much as from 30% to 100% [10]. While marker testing using formalin-fixed, paraffin-embedded there are numerous reports on the sensitivity and speci- (FFPE) samples remains the reference standard, yet ficity of different platforms using tissue as a gold stan- this approach may be limited by the availability of dard [11,12], studies evaluating the agreement between tumor and quality of DNA. Conversely, there is con- different methods for liquid biopsy analysis in large siderable evidence demonstrating that cell-free DNA prospective cohorts remain limited, and reports compar- (cfDNA) genotyping represents a viable and faster ing quantitative measurements of mutant allele frequen- approach [7,8]. In this way, cfDNA testing is recom- cies (MAFs) are particularly scarce. mended for guiding first- and second-line treatment in In this study, we describe the results of an observa- specific clinical circumstances, most notably, when a tional trial specifically designed to evaluate the agree- patient is medically unfit for invasive tissue sampling ment between seven methods for the detection of EGFR or when following pathologic confirmation of a mutations in blood, including two next-generation NSCLC diagnosis, there is no sufficient material for sequencing (NGS)-based methods, three high-sensitivity molecular testing. Indeed, guidelines recommend test- PCR-based platforms, and two FDA-approved meth- ing the T790M resistance mutation in the blood after ods. To our knowledge, this is the largest study (in terms progression to an EGFR TKI as a first choice, and of the number of platforms evaluated) so far to formally rebiopsies are suggested in case of a negative result [9]. compare available technologies designed for noninvasive As a result, EGFR genotyping using plasma samples is EGFR mutation testing and the first to comprehensively becoming widely used in the clinical setting. However, evaluate the concordance between MAFs. In addition, EGFR mutation detection in plasma samples is subject causes for discordant results are thoroughly investigated. 44 Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License A. Romero et al. Liquid biopsy-based cross-platform comparison 2. Methods 2.2. Laboratory procedures Plasma was separated from the cellular fraction by two consecutive centrifugations at 1600 g for 10 min 2.1. Study cohort and at 6000 g for 10 min. Samples were then divided This is a non-post-authorization (non-PAS) and nonin- into 6 aliquots of 2.0 mL and stored at 80 °C, until terventional prospective, multicentre, cross-sectional further analysis or distribution to L2 and L3. study (ClinicalTrials.gov Identifier: NCT03363139) in which 72 patients with EGFR-mutant NSCLC were 2.3. DNA extraction enrolled between July 2018 and January 2019 in 23 Spanish hospitals. There was no intention for vali- In order to compare cfDNA yield between different dation of any technique in the study. The study proto- extraction methods, all samples were processed with col was approved by the Hospital Puerta de Hierro the following methods: (a) Maxwell RSC (MR) Ethics Committee (internal code PI-154/17), and ccfDNA Plasma Kit (Promega Corporation, Madison, informed consent was obtained from all participants. WI, USA), (b) QIAamp Circulating Nucleic Acid Study monitoring was carried out by the Spanish Lung (QCNA) (QIAgen, Valencia, CA, USA), and (c) Cancer Group (www.gecp.org). Inclusion criteria were QIAsymphony DSP Virus/Pathogen Midi Kit using a as follows: (a) patients diagnosed with EGFR-mutant, QIAsymphony (QS) robot (QIAgen), in all cases fol- stage IIIB and IV non-small-cell lung cancer, who lowing the manufacturer’s instructions. The input vol- have progressed as assessed by CT scan according to ume as well as the final elution volume per method is RECIST criteria v.1.1 to first- or second-generation presented in Table S1. cfDNA concentration was mea- EGFR tyrosine kinase inhibitors (TKIs) (e.g., gefitinib, sured using the Qubit dsDNA HS Assay kit (Thermo erlotinib, afatinib), including patients who received a Fisher Scientific, Waltham, MA, USA) on a Qubit 2.0 chemotherapy line before TKI treatment. Samples Fluorometer (Thermo Fisher Scientific). must be drawn before the patient starts a new treat- ment. (b) Patients must sign the informed consent of 2.4. cfDNA genotyping the study. (c) Patients aged ≥ 18 years. Exclusion crite- ria were as follows: (a) patients progressing to third- The presence of EGFR mutations in the purified generation EGFR TKIs (e.g., osimertinib) and (b) no cfDNA was evaluated in the 72 samples using 7 differ- possibility of venipuncture. The patients participating ent methods, namely cobas EGFR Mutation Test v2 in this noninterventional study did not receive treat- (Roche Diagnostics, Penzberg, Germany), Therascreen ment in relation to the study. Prospective information EGFR Plasma RGQ PCR Kit (QIAgen), OncoBEAM about treatments was not collected. EGFR (Sysmex Inostics, Hamburg, Germany), Quant- Three blood samples blood samples were collected per Studio 3D Digital PCR System (Applied Biosystems, patient upon disease progression after a first-line treat- South San Francisco, CA, USA), a 50 -nuclease real-time ment with a first- or second-generation TKI, assessed by PCR (TaqMan, Thermo Fisher Scientific) assay in RECIST criteria v. 1.1, and before the initiation of sec- presence of a peptic nucleic acid probe (PNA-Q-PCR), ond-line treatment. Median time between the assessment and two NGS platforms (Ion S5TM XL and GeneReadTM) of progressive disease and blood extraction was below using two different gene panels, OncomineTM Pan-Cancer 4 weeks in all cases. Eligible patients were both male and Cell-Free Assay (Thermo Fisher, Palo Alto, CA, USA) female, age > 18 years, with a pathologically confirmed and QIAact Lung DNA UMI Panel (QIAgen). Figure S1 diagnosis of stage IV NSCLC harboring an EGFR muta- shows the study flowchart, indicating which methods tion, and who had progressed with first-line EGFR TKI were used in each of the three laboratories. The limit of treatment. In all cases, whole-blood samples were col- detection (LOD) of each method in terms of MAF is pre- lected in a 10-mL Streck Cell-Free DNA BCT tube sented in Table S2. (Streck, La Vista, NE, USA) and in two 8.5-mL PPTTM tubes (Becton Dickinson, Franklin Lakes, NJ, USA). 2.4.1. FDA-approved methods Samples were first sent to a central laboratory 1 (L1) for storage, processing, and distribution of the plasma sam- Cobas EGFR Mutation Test v2 (Roche Diagnostics) ples to the other two central laboratories (L2 and L3). In and Therascreen EGFR Plasma RGQ PCR Kit (QIA- all cases, the oncologist was blinded to laboratory result. gen) were used according to the instructions of the L1, L2, and L3 were blinded for EGFR mutation status. manufacturers. Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 45 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Liquid biopsy-based cross-platform comparison A. Romero et al. 2249del15 c.2236_2250del15; p.L747-S753>S c.2240_ 2.4.2. OncoBEAMTM EGFR kit (Sysmex) 2257del18; p.L858R c.2573T>G 2573_2574TG>GT; BEAMing is a highly sensitive and quantitative plat- p.G719A c.2156G>C; p.T790M c.2369C>T; p.G719S form based on multiplex PCR (mPCR) targeting c.2155G>A; p.G719C c.2155G>T; p.L747_T751delL- somatic alterations and followed by a second PCR REAT c.2239_2253del15; p.L747_S752delLREATS amplification performed on magnetic beads compart- c.2239_2256del18; p.L861Q c.2582T>A; p.C797S mentalized in millions of oil emulsions). All experi- c.2390G>C; 2389T>A; p.E746-S752>V c.2237_2255>T; ments were performed according to the supplier’s p.L747_P753>Q c.2239_2258>CA. recommendations. Briefly, 125 lL of cfDNA samples were employed for mPCR. After mPCR, six replicates 2.4.4. PNA-Q-PCR of each sample were mixed in another plate to perform nested-PCR. Then, serial dilutions (1 : 50/1 : 60/1 : 50) The assay is based on quantitative real-time PCR were carried out. The diluted samples were then trans- (TaqMan) in the presence of a PNA clamp (Eurogen- ferred to the emulsion PCR plate, together with emul- tec, Seraing, Belgium) designed to inhibit the amplifi- sion working reagents (one for each codon). After this cation of the wild-type alleles. The assay has been fully step, the EmulsiFIRE solution was added to induce validated, has an ISO15189 accreditation, and allows the emulsion, creating millions of PCR compartments estimation of the absolute and relative abundances of (hydrophilic droplets with a single magnetic bead mutant alleles in positive samples. Briefly, amplifica- inside). After breaking the emulsion PCR, WT and tion is performed in a final volume of 12.5 lL, using mutant-specific probes were hybridized and analyzed 3 lL (~ 4.5 ng) for exon 21 analysis or 1 lL by flow cytometry using the Cube 6i cytometer of Sys- (~ 1.5 ng) for exon 19 and p.T790M analysis of mex (Sysmex Inostics). Plasma was considered posi- cfDNA 6.25 lL of Genotyping Master Mix (Applied tive by BEAMing for a given mutation if the mutation Biosystems, Foster City, CA, USA) 0.96 pmol of each was detected above thresholds used for clinical appli- primer 1.2 pmol of probes and 6.25 pmol (for exon 21 cation (Table S2). and p.T790M) or 62.4 pmol (for exon 19) of PNA. Samples are submitted to 50 cycles of 15 s at 92 °C and 1.5 min at 60 °C, in a QuantStudioTM 6 real-time 2.4.3. QuantStudio 3D Digital PCR (dPCR) PCR System (Applied Biosystems/Thermo Fisher Sci- Cell-free DNA was analyzed using commercially avail- entific). The sequence of the primers, probes, and able predesigned TaqMan Liquid Biopsy dPCR PNAs used and analytical performance has been assays on a QuantStudio 3D Digital PCR System described elsewhere [10,14]. Specifically, the assay cov- (Applied Biosystems). The dPCR was carried out on a ers the following mutations: p.L747-T751 c.2238_ final volume of 18 lL and using 8.55 lL of cfDNA 2252del15; p.L747_S753>S c.2240_2257del18; p.E746_ template. Subsequently, 14.5 lL was loaded into a A750delELREA c.2235_2249del15, c.2236_2250del15; QuantStudio 3D Digital PCR 20K chip. The cycling p.L747_S752delLREATS c.2239_2256del18; p.E746_ conditions were as follows, initial denaturation at S752>V c.2237_2255>T; p.E746_T751>A c.2237_ 96 °C for 10 min, followed by 40 cycles at 56 °C for 2244>T; p.L747_T751>P c.2239_2251>C; p.L747_ 2 min, and at 98 °C for 30 s, a step of 72 °C for A750>P c.2239_2248TTAAGAGAAG>C; p.L858R 10 min, and finally, samples were maintained at 22 °C c.2573T>G 2573_2574TG>GT; p.G719A c.2156G>C; for at least 30 min. Chip fluorescence was read twice. p.T790M c.2369C>T; p.G719S c.2155G>A; p.G719C Results were analyzed with QUANTSTUDIO 3D ANALYSIS c.2155G>T; p.L861Q c.2582T>A; p.C797S c.2390G>C. TM SUITE CLOUD Software (Thermo Fisher Scientific). Analyses were carried out in duplicate using one sam- The automatic call assignments for each data cluster ple of purified cfDNA, when possible. In addition, all were manually adjusted when needed. The result of the samples were assayed in the absence of PNA to con- assay is reported as the ratio of mutant DNA mole- firm the presence of cfDNA. Genomic DNAs from cell cules relative to the sum of mutant and wild-type (wt) lines at 1.5 nglL1 were used as positive and negative DNA molecules. A negative control DNA was controls. Extraction and nontemplate controls were included in every run. Details about assay performance added in each run. A sample was considered positive if have been described elsewhere [13]. Mutations detected the same mutant allele amplified in the two duplicates by this assay include p.L747_T751>P c.2239_2251>C; in the presence of PNA. If amplification was only p.L747_A750>P c.2238_2248>GC; p.E746_T751delEL- detected in one duplicate, samples were reanalyzed and REAT c.2236_2253del18; p.L747_T751delLREAT considered positive if again at least one of the dupli- c.2238_2252del15; p.E746_A750delELREA c.2235_ cates was positive for the same mutated allele. 46 Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License A. Romero et al. Liquid biopsy-based cross-platform comparison bead enrichment, pooled libraries were sequenced 2.4.5. NGS with the oncomine pan-cancer cell-free using the GeneRead UMI Advanced Sequencing Q kit assay (Oncomine) in a GeneReader instrument. QIAGEN CLINICAL INSIGHT Library preparation was performed according to ANALYZE software (QIAgen) was employed to perform manufacturer’s instructions. All the purifications were the secondary analysis of FASTQ reads, align the read made using AMPure XP magnetic beads (Beckman data to the hg19 reference genome sequence, call Coulter, Brea, CA, USA). Library quantification was sequence variants, and generate a report for visualiza- performed using Ion Library TaqMan Quantitation tion of the sequencing results. Variants were imported kit (Thermo Fisher) in a StepOnePlusTM qPCR into the QIAGEN Clinical Insight Interpret web inter- machine (Thermo Fisher). The individual libraries face for data interpretation and generation of final cus- were diluted to a final concentration of 100 pM. The tom report. final barcoded libraries were pooled and adjusted to a final concentration of 50 pM. Template preparation 2.5. Statistical analysis and chip loading were carried out on an Ion ChefTM System (Thermo Fisher). Eight samples were loaded For the analysis of cfDNA extraction yield according onto an Ion 550TM chip. Finally, Ion 550TM chips were to extraction method, a Friedman test was carried sequenced on an Ion S5TM Sequencer (Thermo out to assess whether the measurements of the Fisher). Analysis of raw sequencing data was per- amount of cfDNA (ng) normalized by milliliter of formed using TORRENT SUITE Software (v5.10.0, plasma obtained by each methodology were equiva- Thermo Fisher). For sequencing coverage analysis, lent. The Friedman test is used for one-way repeated- the COVERAGEANALYSIS (v.5.10.0.3) plug-in was used measures analysis of variance by ranks [15]. The dif- (Thermo Fisher). Raw reads were aligned to the ference in AFs between EGFR-sensitizing and human reference genome hg19. Variant calling, anno- p.T790M mutation was assessed using the Wilcoxon tation, and filtering were performed on the Ion signed-rank test. The agreement between different Reporter (v5.10) platform using the Oncomine Taq- methodologies for the assessment of p.T790M status Seq Pan-Cancer Liquid Biopsy workflow (v5.10). The as well as the original EGFR-sensitizing mutation clinical significance of somatic variants was performed status (detected vs not detected) was evaluated using according to Standards and Guidelines for the Inter- the kappa coefficient values and the corresponding pretation and Reporting of Sequence Variants in 95% confidence intervals (95% CI). The strength of Cancer. agreement is considered slight when the K values are between 0.00 and 0.20 fair, 0.21 and 0.40 moderate, 0.41 and 0.60 good, 0.61 and 0.80 and almost perfect, 2.4.6. NGS with the GeneReader platform 0.81 and 0.99 [16]. To evaluate the relative accuracy (GeneReader) of each method, EGFR mutation status using the Purified DNA (16.75 lL, ~ 10–70 ng) was used as a FFPE sample from the rebiopsy was considered the template to generate libraries for sequencing with the gold standard. For this purpose, sensitivity, speci- GeneRead TM QIAact Lung DNA UMI Panel, ficity, positive likelihood ratio, and negative likeli- according to manufacturer’s instructions. The panel is hood ratio were calculated. designed to enrich specific target regions containing The difference in MAFs between EGFR mutations 550 variant positions in 19 selected genes frequently was assessed using the Wilcoxon signed-rank test and altered in lung cancer tumors (AKT1, ALK, BRAF, the reliability between MAFs measured by different DDR2, EGFR, ERBB2/HER2, ESR1, KIT, KRAS, methodologies was evaluated using the intraclass cor- MAP2K1, MET, NRAS, NTRK1, PDGFRA, relation coefficient (ICC) through a two-way mixed-ef- PIK3CA, PTEN, ROS1, FGFR1, and RICTOR), fects model, along with the 95% confidence interval. including MET exon 14 skipping mutations. Libraries Lin’s concordance correlation coefficient [17,18] was were quantified using a QIAxcel Advanced System estimated to assess the concordance between the mea- (QIAgen) and Qubit dsDNA HS Assay kit (Thermo sures two by two, interpreting the coefficient as poor if Fisher Scientific), diluted to 100 pg/lL, and pooled in < 0.90, 0.90–0.95 as moderate, 0.95–0.99 as substantial batches of 6 (liquid biopsies). Clonal amplification was and > 0.99 almost perfect [19]. performed on 625 pg of pooled libraries by the Gene- In addition, Passing and Bablok [20] regression anal- Read Clonal Amp Q Kit (QIAgen) using the Gene- ysis was performed to assess the agreement and possi- Read QIAcube and an automated protocol. Following ble systematic bias between methods. Linear model Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 47 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Liquid biopsy-based cross-platform comparison A. Romero et al. validity was performed using the CUSUM test for lin- Table 1. Clinicopathological characteristics of the study population. earity. Similarly, AF measurements by NGS-based Feature Value % methodologies employed were graphically displayed by means of Bland–Altman plot [21]. Age (years) 66 Sex (N) Statistical significance was set at P < 0.05. The sta- Male 24 33 tistical analysis was performed using STATA v.15.1 soft- Female 48 67 ware (StataCorp. 2017; Stata Statistical Software: Smoking status (N) Release 15, College Station, TX, USA: StataCorp Current/ex 30 42 LLC.) Never 42 58 Histology (N) Adenocarcinoma 66 92 3. Results Other 6 8 Stage IVA 50 69 3.1. Study cohort IVB 22 31 EGFR mutation (N) Clinico-pathological characteristics of the study popu- Deletion exon 19 35 49 lation are presented in Table 1. Most patients were Point mutation exon 21 33 46 women (66.67%), of Caucasian ethnicity (95.83%), age Other 4 6 Treatment range at diagnosis from 37 to 84 years, with an aver- Afatinib 20 28 age of 65.5 years, and 58.33% of the patients were Erlotinib 19 26 never smokers. According to the pathology reports, Gefitinib 31 43 91.67% of the cases were adenocarcinomas; the rest of Other 2 3 the cases (8.33%) corresponded to adenosquamous and large-cell carcinoma. All patients were stage IV, with 69% being stage IVA. Regarding EGFR status, 48.61% harbored a deletion in exon 19, 45.83% a point mutation in exon 21, 2.78% a point mutation in 3.3. Agreement between methods for EGFR exon 18, one tumor harbored S768I in exon 20, and mutation detection another tumor harbored an insertion in exon 20. There Purified cfDNA samples were analyzed in three central were no significant differences in mutation distribution laboratories using a total of seven methods. The pro- according gender, age, or histology. Finally, the med- portion of observed agreement and Cohen’s kappa ian number of metastatic lesions at disease progression index (K) between methods is shown in Table 3. The was 3 (range 1–12), and 27.8% of patients had central agreement between all methods was almost perfect for nervous system (CNS) involvement. the detection of deletions in exon 19 (K = 0.87; 95% CI: 0.78–0.96) and substantial for exon 21 point 3.2. Comparison of DNA extraction methods mutations (K = 0.76; 95% CI: 0.63–0.89). Regarding the T790M resistance mutation, concordance was Plasma samples were aliquoted and used for cfDNA lower but still substantial (K = 0.68; 95% CI: 0.57– extraction using three different methods: MR, QCNA, 0.79). The comparison of two FDA-approved methods and QS. There was a moderate to strong correlation (cobas and Therascreen) showed almost perfect between the amount of cfDNA (ngmL1 plasma) agreement for the detection of exon 19 deletions and obtained by each methodology (Pearson correlation T790M mutation (K = 0.81; 95% CI: 0.65–0.97 and coefficients: 0.85, 0.89, and 0.98 for QCNA-MR, MR- QS, and QCNA-QS comparisons, respectively; P < 0.05 in all cases). The median total amounts of Table 2. Extraction yield. Median, P25 and P75 of the amount (ng) cfDNA (ng) normalized by plasma input volume, as of cfDNA obtained per mL of plasma. well as the P25 and P75 yielded by each method, are presented in Table 2. There were significant differences Extraction method in the cfDNA isolation yield between the methods QCNA MR QS evaluated (P < 0.001), being lower for MR method Median 19.71 12.25 18.25 compared to QCNA and QS. Total amount of cfDNA P25 13.91 6.78 13.47 obtained for each patient and according to each P75 28.68 24.06 25.50 methodology is presented in Table S3. 48 Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License A. Romero et al. Liquid biopsy-based cross-platform comparison K = 0.83; 95% CI: 0.66–1.00, respectively) and sub- perfect for the detection of exon 19 deletions and the stantial agreement for the identification of point muta- T790M mutation and increased to K = 0.93 (95% CI: tions in exon 21 (K = 0.61; 95% CI: 0.40–0.82). 0.83–1.00) in the case of point mutations in exon 21 Remarkably, the agreement between NGS platforms (Table S4). The agreement for exon 21 and T790M for the detection of EGFR-activating mutations as well mutations also improved in the case of NGS-based as the T790M mutation was particularly good (Fig. 1) methods when samples with MAFs ≤ 0.5% were (K = 0.84; 95% CI: 0.68–1.00, K = 0.83; 95% CI: excluded from the analysis. 0.66–1.00 and K = 0.77; 95% CI: 0.59–0.95 for the Finally, we investigated whether any clinico-patho- detection of mutations in exon 19, 21, and T790M, logical feature of the patients was associated with respectively). Finally, high-sensitivity PCR-based lower MAFs and could be a potential indicator of methodologies showed perfect to substantial agreement tumor shedding. In this study, ctDNA levels were not (K = 0.89; 95% CI: 0.78–0.99, K = 0.80; 95% CI: dependent on age, sex, histology, sum of metastatic 0.67–0.92 and K = 0.68; 95% CI: 0.54–0.83 for the lesions, or metastasis location, except for the presence detection of mutations in exon 19, 21, and T790M, of CNS metastasis. EGFR-sensitizing mutations were respectively). detected at significantly lower MAFs in patients pro- gressing exclusively at the CNS level compared to patients with disease progression assessed at other 3.4. Investigation of discordant results anatomical locations (Fig. S2). The T790M detection The possible causes of discordant results were investi- rate was also lower in the subset of patients progress- gated by elucidating the relationship between the ing at CNS exclusively (Table S5). detection of EGFR mutations and cfDNA input (ng) and time between assessment of progressive disease 3.5. Correlation between MAFs obtained by and blood drawn, but no significant differences in different methods cfDNA input or withdrawn timing were found between concordant and discordant samples. Next, we In our patient cohort, MAFs of positive samples ran- examined whether discordant results were related to ged from 0.02% to 63.9%. A list containing all differences in the limits of detection (LOD) of the dif- detected mutations and corresponding MAFs is avail- ferent assays and, consequently, if discordant calls able in Data S1. We assessed the reliability of MAFs occurred mostly at low MAFs. We thus performed an obtained using quantitative techniques by first calculat- agreement analysis discarding samples in which EGFR ing the intraclass correlation coefficients (ICCs). As mutations were detected at MAFs ≤ 0.5%. It is note- shown in Table 4, the MAF results determined by the worthy that, in this subset of samples, the agreement two NGS-based methods, OncomineTM Pan-Cancer between high-sensitivity PCR-based methods was Cell-Free Assay and GeneReadTM QIAact Lung DNA UMI Cancer Panel, were almost identical (ICC = 0.98; 95% CI: 0.96–0.99 for EGFR-sensitizing mutations Table 3. Agreement between methodologies for the detection of and ICC = 0.97; 95% CI: 0.95–0.98 for T790M). Simi- deletions in exon 19, point mutations in exon 21 and the T790M larly, MAFs estimated using high-sensitivity PCR- mutation. Percentage of agreement, Cohen’s kappa index (K), and based platforms showed an excellent agreement corresponding confidence intervals. (ICC = 0.93; 95% CI: 0.90–0.96 for EGFR-sensitizing Comparison group Mutation Agreement Kappa 95% CI mutations and ICC = 0.86; 95% CI: 0.79–0.91 for T790M). All techniques Exon 19 95.50 0.87 0.78–0.96 Next, we used Passing–Bablok regression analysis to Exon 21 91.26 0.76 0.63–0.89 estimate the agreement and possible systematic bias T790M 86.83 0.68 0.57–0.79 between NGS-based methods and obtained equations IVD-approved Exon 19 92.86 0.81 0.65–0.97 Exon 21 85.71 0.61 0.40–0.82 of Y = 1.107 (95% CI: 1.029–1.200) X for EGFR-acti- T790M 94.29 0.83 0.66–1.0 vating mutations and Y = 1.059 (95% CI: 0.974– High-sensitivity PCR Exon 19 96.15 0.89 0.78–0.99 1.585) X for MAFs of the T790M mutation (Fig. 2A). based Exon 21 91.67 0.80 0.67–0.92 A CUSUM test indicated no significant deviation from T790M 85.19 0.68 0.54–0.83 linearity (P > 0.20), while the Bland–Altman plot NGS-based Exon 19 94.37 0.84 0.68–1.00 (Fig. 2B) showed little bias between the two NGS- Exon 21 94.2 0.83 0.66–1.00 based methods for EGFR-sensitizing mutation and T790M 91.55 0.77 0.59–0.95 T790M MAF quantification (bias = 0.85, 95- % CI = 12.99 to 14.69 and bias = 0.013, Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 49 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Liquid biopsy-based cross-platform comparison A. Romero et al. Fig. 1. Venn diagrams showing concordance among NGS-based methodologies and PCR-based platforms for T790M detection. L2 NGS: OncomineTM Pan-Cancer Cell-Free Assay performed in laboratory 2. L3 NGS: GeneReadTM QIAact Lung DNA UMI Cancer Panel performed in laboratory 3. L2 dPCR: QuantStudio3D Digital PCR System, performed in laboratory 2. L3 TaqMan in-house 5-nuclease real-time PCR assay in presence of PNA carried out in laboratory 3. L1 BEAMing OncoBEAMEGFRperformed in laboratory 1. Table 4. ICC and corresponding confidence intervals for equivalent with any clinic-pathological features analyzed (gender, MAF measurements. age, histology, smoking status). Comparison group Mutation ICC 95% CI The sensitivity, specificity, and positive and negative likelihood ratios for each method, considering the All techniques Sensitizing 0.94 0.92–0.96 T790M 0.94 0.91–0.96 results of rebiopsy as the gold standard, are shown in High-sensitivity PCR based Sensitizing 0.93 0.90–0.96 Table 5. Of note, the G719X and S768l mutations, T790M 0.86 0.79–0.91 which were reported to be present in two rebiopsies NGS-based Sensitizing 0.98 0.96–0.99 from two independent patients, were not detected by T790M 0.97 0.95–0.98 any of the methodologies. As shown, all parameters were superior for EGFR-sensitizing mutations com- pared with the T790M mutation. Remarkably, the T790M was systematically detected at lower MAF in 95% CI = 3.12–3.14, respectively), with only three the plasma sample than the original EGFR-sensitizing measurements lying outside the confidence interval in mutation (P < 0.001; Wilcoxon test) (Fig. S4), suggest- both cases. Similar results were obtained for PCR- ing that the lower MAF values for T790M could limit based platforms (Fig. S3). assay performance. Finally, there was a significant but weak positive Sensitivity and specificity of each methodology, for correlation between MAFs and total cfDNA concen- the detection of the original EGFR sensitizing consid- tration (Spearman’s correlation coefficient below 0.4 in ering the gold standard the original tissue specimen is all cases). presented in Table S6. 3.6. Comparison with tissue biopsies 4. Discussion A rebiopsy at disease progression was obtained from 35 Biomarker testing using liquid biopsies is becoming (48.61%) patients. According to the pathology report, the standard of care in many clinical laboratories as a the original EGFR-sensitizing mutation was detected in noninvasive procedure with a short turnaround time. all FFPE samples, while the T790M mutation was pre- However, studies assessing the agreement between dif- sent in 15 tumor samples (43%). The T790M mutation ferent platforms and, more importantly, exploring bio- was more frequent in patients whose tumors harbored a logical and technical factors responsible for discordant deletion in exon 19 (77%) than in those with other sen- results are still limited. Here, we present the results of sitizing mutations (54%). Of note, in cases in which an observational trial specifically designed to evaluate rebiopsy was not possible (N = 37), the T790M muta- the agreement between seven methods used for the tion was detected in the plasma sample by at least one detection of EGFR mutations in the cfDNA. Samples method in 18 cases. Overall, T790M mutation was from 72 patients with stage IV NSCLC progressing to detected by at least one method in 39 (54%) plasma a first- or second-generation TKI were prospectively samples. The detection of the T790M mutation in the collected in 23 Spanish hospitals under a strict proto- plasma was not associated with the TKI received nor col. The study was monitored by the Spanish Lung 50 Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License A. Romero et al. Liquid biopsy-based cross-platform comparison Fig. 2. Comparison of MAFs obtained by NGS-based platforms. (A) Passing-Bablok regression showing close concordance between the two methods for the assessment of T790M MAFs and Bland–Altman plot showing low level of bias between both methods for quantifying T790M allele frequency. (B) Passing–Bablok regression and Bland–Altman plot showing the agreement between NGS-based platforms for the quantification ofEGFR-sensitizing mutations. Table 5. Sensitivity, specificity positive predictive value (PPV) and negative predictive value (NPV) of each methodology and according to the type of mutation. T790M Exon 21 Exon 19 Sensitivity Specificity PPV NNV Sensitivity Specificity PPV NNV Sensitivity Specificity PPV NNV Methodology (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) BEAMing 54 73 81 42 80 100 100 92 71 100 100 70 dPCR 58 73 82 44 90 100 100 96 62 100 100 64 PNA-Q-PCR 54 73 81 42 70 100 100 89 53 100 100 58 NGS 42 90 91 40 89 100 100 96 62 92 93 60 Oncomine NGS 42 82 83 40 70 100 100 89 61 100 100 57 GeneReader Cobas 42 91 91 42 80 100 100 92 65 93 100 65 Therascreen 25 90 86 33 50 100 100 82 50 100 100 58 Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 51 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Liquid biopsy-based cross-platform comparison A. Romero et al. Cancer Group (GECP) to minimize the variability uncommon clinically relevant EGFR mutations in due to preanalytical conditions. Overall, the propor- exons 18, 19, 20, 21 might not be detected by qPCR- tion of observed agreement among all the methods based technologies. In this way, singleplex approaches for the detection of EGFR mutations was good and such as dPCR are limited by the number of mutations particularly high for NGS-based and high-sensitivity that can be interrogated in a given sample and might PCR platforms. Interestingly, concordance was require to prioritize the order of mutations to be slightly lower for T790M mutation compared to tested. On the contrary, NGS-based approaches permit mutations in exons 19 and 21, which always presented to interrogate a wide number of mutations simultane- higher MAFs (P < 0.001; Wilcoxon test) (Fig. S4), ously saving time and sample material. consistent with the idea that T790M mutation is sub- We explicitly examined the effect of the input quan- clonal and arises later in tumor evolution and suggest- tity DNA and MAF for discordant calls. First, we ing that the lower MAF values for T790M mutation were not able to demonstrate an improvement in the could limit assay performance. A cross-platform com- agreement between methods when samples with low parison including 38 samples from patients with cfDNA input were discarded from the analysis. Con- EGFR-mutated lung cancer from the phase 1 AURA versely, the agreement ranged from almost perfect to trial analyzed the concordance between the results perfect (K = 1) when excluding samples in which an obtained with BEAMing, ddPCR, Therascreen, and EGFR mutation was detected at MAFs ≤ 0.5%. This cobas EGFR Mutation Test using tissue results as a observation could also explain the higher agreement nonreference standard [22]. Consistent with our and better sensitivity and specificity of methods when results, the authors found that the concordance was analyzing EGFR-sensitizing mutation compared to the lower for T790M (57%, 48%, 74%, and 70% for concomitant T790M mutation. As mentioned, the T790M vs 97%, 95%, 97%, and 95% for the L858R T790M mutation was always detected at lower MAFs mutation using cobas, Therascreen, ddPCR, and with respect to the original EGFR-sensitizing mutation. BEAMing, respectively). In a recent study analyzing Consistent with this, in a recent paper comparing EGFR mutations in plasma from patients recruited in BEAMing and ddPCR for ctDNA analysis using the AURA3 trial using the cobas EGFR Mutation plasma samples from advanced breast cancer patients Test v2 (cobas plasma), droplet digital polymerase enrolled in the PALOMA-3 trial, the authors showed chain reaction (ddPCR), and an NGS-based test that discordant calls occurred at MAFs < 1% [24]. (Guardant360), a lower positive percent agreement Likewise, in the above-mentioned study analyzing sam- was observed for the detection of T790M mutation ples from the AURA3 trial, the authors reported that compared with EGFR exon 19 deletion or L858R discordant results between NGS and ddPCR always mutation [23]. occurred in cases where mutations were detected at In our study, rebiopsies were obtained from 35 MAFs ≤ 1% [23]. Nevertheless, this study did not (48.61%) patients, highlighting the limited tissue avail- explore the patient’s outcome according to MAFs, as ability after disease progression due to patient safety it was conceived a non-post-authorization (non-PAS), concerns. Of note, the T790M mutation was detected noninterventional study where participants did not in 18 cases in which rebiopsy was not feasible. receive treatment in relation to the study. Further Considering tissue genotyping as a nonreference studies addressing the impact on MAF in treatment standard, in our study sensitivity ranged from 25% to outcome would be of particular interest. 58% for the detection of T790M mutation. Consistent We were not able to demonstrate a significant asso- with our results, T790M mutation detection rates in ciation between DNA input and discordant results, blood samples collected upon disease progression have but the opposite hypothesis cannot be ruled out. been reported to range as broadly as from 18% to Indeed, there was a significant but weak positive corre- 78% [10]. Specificity, however, was acceptable for real- lation between MAFs and total cfDNA concentration. world applications, which supports the use of blood as Other researchers have previously reported this obser- a first choice for assessing EGFR mutations and rele- vation [25], meaning that EGFR mutations can be gating tissue tests to cases of negative plasma results. detected in the cfDNA at AF > 0.5%, even when the NGS, dPCR, PNA-Q-PCR, and BEAMing detected a amount of cfDNA is low. Therefore, in our hands, higher number of T790M-positive samples than cobas MAFs is the best parameter for evaluating how trust- and Therascreen as was observed previously, but with worthy the result of a plasma test is and should be fewer number or cross comparisons than in our study always informed in clinical reports. Clinical guidelines [10,22,23]. It is important to point out that EGFR-sen- should stress the fact that plasma genotyping is less sitizing mutations landscape is complex and informative when MAFs are missing. 52 Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License A. Romero et al. Liquid biopsy-based cross-platform comparison Despite emerging evidence suggesting that quantita- of other clinical factors affecting tumor shedding (i.e., tive information from plasma genotyping is clinically tumor stage). relevant [24,25], the available data evaluating the con- The strengths of our study include the comparison cordance between quantitative MAF values are still of many platforms, in contrast with previously pub- limited to small sets. Here, we compared MAF mea- lished studies, sample anonymization and blind analy- surements assessed by each method using statistical sis by the participating laboratories and, finally, that methods common in laboratory medicine, namely the samples were prospectively collected under a strict pro- Passing-Bablok regression and Bland–Altman plots. tocol, with all enrolled patients being at the same clini- However, these methods are scarcely used for compar- cal treatment stage, immediately after progression to a ing the performance of molecular biology-based tech- first-line TKI treatment. Moreover, the study was niques [26,27]. Nevertheless, we believe that MAFs can monitored by a contract research organization (Span- be considered as a clinical chemistry analyte, such as ish Lung Cancer Group) to minimize clinical variabil- glucose or cholesterol, and therefore, methods should ity. One limitation is that we did not measure be compared using statistical approaches specifically potential contamination from somatic mutations attri- designed for this aim. Importantly, our data indicate butable to clonal hematopoiesis, although, to our that there was an excellent concordance between knowledge, somatic mutations in the EGFR gene due MAFs obtained by NGS-based techniques and to clonal hematopoiesis are very rare events at best obtained by high-sensitivity PCR-based methods. With [32,33]. an analytic sensitivity like that of dPCR or BEAMing, NGS-based assay would be the best option for study- 5. Conclusion ing oncogenic drivers in the ctDNA as a wider number of somatic mutations can be interrogated. A limited This prospective multicenter study demonstrates that number of studies have evaluated the concordance NGS, digital PCR, and RT-PCR-based methodologies between NGS-based assays using clinical samples and show good to excellent agreement for the detection of conflicting results have been reported. While some EGFR mutations in cfDNA, including the T790M researchers have found very low congruence between mutation, with most discordant calls occurring at platforms for same-patient-paired samples [28], others MAFs ≤ 0.5%. With NGS enabling the simultaneous have reported a high rate of concordance in direct testing of multiple mutations, our results support the comparisons between NGS-based platforms, with dis- use of this technology for noninvasive biomarker test- cordant somatic mutations being mostly subclonal ing and suggest that MAFs and the limits of detection [29]. This suggests that discordant calls mostly occur of the assay used should always be reported in the at low MAF values. clinical setting. Driver mutation MAFs varied from patient to patient by as much as 0.02% to 54.4%. To elucidate Acknowledgements which factors could determine such a difference, we evaluated several clinical variables. In our hands, the We would like to thank our patients for their partici- presence of CNS metastasis was the only factor affect- pation in the study. ing MAFs, suggesting that biomarker testing using liq- uid biopsies could be limited by the anatomical Conflict of interest location of the cancer lesions and that assay sensitivity might be compromised in patients with CNS metasta- The authors have declared no conflict of interest at sis. Indeed, it is well established that ctDNA is more individual level. AstraZeneca supported this work. All frequently detected in patients with solid tumors out- reagents, necessary for the study, were funded by side the brain [30], and we have recently demonstrated AstraZeneca. The study protocol and the final version that pleural effusion, ascites, and cerebrospinal fluid of the manuscript were revised and approved by Astra- are superior to blood for detecting somatic mutations Zenca. in patients with pleural, peritoneal, or CNS involve- ment [31]. On the other hand, since our cohort was Author contributions very homogenous in terms of the type of patients included (i.e., all patients had stage IV disease, samples AR, EJ-L, MAM-V conceived and coordinated the were all collected at the first disease progression after study. AR, EJ-L, MAM-V, MP, BG-P, ARoy, SC-F, first-line treatment with TKI, most cases were adeno- AR-F, and MP contributed to experimental design carcinoma, etc.), we were unable to analyze the impact and data analysis. SC-F, AR-F, BG-P, AI, PC, and Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 53 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Liquid biopsy-based cross-platform comparison A. Romero et al. AC carried out experiments. AR, ARoy, EJ-L, and F, Nogami N, Kurata T et al. (2018) Osimertinib in MAM-V performed the statistical analyses. JPA, OJV, untreated EGFR-mutated advanced non-small-cell lung PD, MC, BH, SVE, GB, MG, MM, RB, ALO, AB, cancer. N Engl J Med 378, 113–125. JB-B, JMJ, JGG, SV, CGG, BM, and ALM selected 7 Rolfo C, Mack PC, Scagliotti GV, Baas P, Barlesi F, the study population. AR, MAM-V, and EJ-L con- Bivona TG, Herbst RS, Mok TS, Peled N, Pirker R tributed to manuscript preparation. All authors read et al. (2018) Liquid biopsy for advanced non-small cell and approved the final manuscript. lung cancer (NSCLC): a statement paper from the IASLC. J Thorac Oncol 13, 1248–1268. 8 Leighl NB, Page RD, Raymond VM, Daniel DB, Ethics approval and consent to Divers SG, Reckamp KL, Villalona-Calero MA, Dix D, participate Odegaard JI, Lanman RB et al. (2019) Clinical utility of comprehensive cell-free DNA analysis to identify The study was approved by the Hospital Puerta de genomic biomarkers in patients with newly diagnosed Hierro Ethics Committee (internal code internal code metastatic non-small cell lung cancer. Clin Cancer Res PI-154/17), and written informed consent was obtained 25, 4691–4700. from each participant. 9 Lung Cancer Metastatic. Available online at NCCN. org/patients NCCN guidelines for patients non-small Data accessibility cell lung cancer. 10 Mayo-de-las-Casas C, Jordana-Ariza N, Garz on-Iba~nez Data are available upon request. M, Balada-Bel A, Bertran-Alamillo J, Viteri-Ramırez S, Reguart N, Mu~ noz-Quintana MA, Lianes-Barragan P, Camps C et al. (2017) Large scale, prospective screening References of EGFR mutations in the blood of advanced NSCLC 1 Sequist LV, Martins RG, Spigel D, Grunberg SM, Spira patients to guide treatment decisions. Ann Oncol 28, A, J€anne PA, Joshi VA, McCollum D, Evans TL, 2248–2255. Muzikansky A et al. (2008) First-line gefitinib in patients 11 Oxnard GR, Paweletz CP, Kuang Y, Mach SL, with advanced non-small-cell lung cancer harboring O’Connell A, Messineo MM, Luke JJ, Butaney M, somatic EGFR mutations. J Clin Oncol 26, 2442–2449. Kirschmeier P, Jackman DM et al. (2014) Noninvasive 2 Mok TS, Wu YL, Thongprasert S, Yang CH, Chu DT, detection of response and resistance in EGFR mutant Saijo N, Sunpaweravong P, Han B, Margono B, lung cancer using quantitative next-generation Ichinose Y et al. (2009) Gefitinib or carboplatin- genotyping of cell-free plasma DNA. Clin Cancer Res paclitaxel in pulmonary adenocarcinoma. N Engl J Med 20, 1698–1705. 361, 947–957. 12 Jenkins S, Yang JCH, Ramalingam SS, Yu K, Patel S, 3 Rosell R, Carcereny E, Gervais R, Vergnenegre A, Weston S, Hodge R, Cantarini M, J€anne PA, Massuti B, Felip E, Palmero R, Garcia-Gomez R, Mitsudomi T et al. (2017) Plasma ctDNA analysis for Pallares C, Sanchez JM et al. (2012) Erlotinib versus detection of the EGFR T790M mutation in patients standard chemotherapy as first-line treatment for with advanced non-small cell lung cancer. J Thorac European patients with advanced EGFR mutation- Oncol 12, 1061–1070. positive non-small-cell lung cancer (EURTAC): a 13 Provencio M, Torrente M, Calvo V, Perez-Callejo D, multicentre, open-label, randomised phase 3 trial. Gutierrez L, Franco F, Perez-Barrios C, Barquın M, Lancet Oncol 13, 239–246. Royuela A, Garcıa-Garcıa F et al. (2018) Prognostic 4 Park K, Tan E-H, O’Byrne K, Zhang L, Boyer M, value of quantitative ctDNA levels in non small cell Mok T, Hirsh V, Yang JC-H, Lee KH, Lu S et al. lung cancer patients. Oncotarget 9, 488–494. (2016) Afatinib versus gefitinib as first-line treatment 14 Karachaliou N, Mayo-De Las Casas C, Queralt C, De of patients with EGFR mutation-positive non-small- Aguirre I, Melloni B, Cardenal F, Garcia-Gomez R, cell lung cancer (LUX-Lung 7): a phase 2B, open- Massuti B, Sanchez JM, Porta R et al. (2015) label, randomised controlled trial. Lancet Oncol 17, Association of EGFR L858R mutation in circulating 577–589. free DNA with survival in the EURTAC trial. JAMA 5 Mok TS, Wu Y-L, Ahn M-J, Garassino MC, Kim HR, Oncol 1, 149–157. Ramalingam SS, Shepherd FA, He Y, Akamatsu H, 15 Friedman M (1937) The use of ranks to avoid the Theelen WS et al. (2017) Osimertinib or platinum- assumption of normality implicit in the analysis of pemetrexed in EGFR T790M–positive lung cancer. N variance. J Am Stat Assoc 32, 675–701. Engl J Med 376, 629–640. 16 Landis JR & Koch GG (1977) The measurement of 6 Soria JC, Ohe Y, Vansteenkiste J, Reungwetwattana T, observer agreement for categorical data. Biometrics 33, Chewaskulyong B, Lee KH, Dechaphunkul A, Imamura 159. 54 Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License A. Romero et al. Liquid biopsy-based cross-platform comparison 17 Lin LI-K (1989) A concordance correlation coefficient 28 Torga G & Pienta KJ (2018) Patient-paired sample to evaluate reproducibility. Biometrics 45, 255. congruence between 2 commercial liquid biopsy tests. 18 Steichen TJ & Cox NJ (2002) A note on the JAMA Oncol 4, 868–870. concordance correlation coefficient. Stata J Promot 29 Gupta R, Othman T, Chen C, Sandhu J, Ouyang C & Commun Stat Stata 2, 183–189. Fakih M (2020) Guardant360 circulating tumor DNA 19 McBride GB (2005) A proposal for strength-of- assay is concordant with FoundationOne next- agreement criteria for Lin’s concordance correlation generation sequencing in detecting actionable driver coefficient. NIWA Client Report, HAM2005-062. mutations in anti-EGFR naive metastatic colorectal 20 Passing H & Bablok W (1983) A new biometrical cancer. Oncologist 25, 235–243. procedure for testing the equality of measurements 30 Bettegowda C, Sausen M, Leary RJ, Kinde I, Wang Y, from two different analytical methods. Application of Agrawal N, Bartlett BR, Wang H, Luber B, Alani RM linear regression procedures for method comparison et al. (2014) Detection of circulating tumor DNA in studies in Clinical Chemistry. Part I. J Clin Chem Clin early- and late-stage human malignancies. Sci Transl Biochem 21, 709–720. Med 6, 224ra24. 21 Martin Bland J & Altman DG (1986) Statistical 31 Villatoro S, Mayo-de-las-Casas C, Jordana-Ariza N, methods for assessing agreement between two methods Viteri-Ramırez S, Garzon-Iba~nez M, Moya-Horno I, of clinical measurement. Lancet 327, 307–310. Garcıa-Pelaez B, Gonzalez-Cao M, Malapelle U, 22 Thress KS, Brant R, Carr TH, Dearden S, Jenkins S, Balada-Bel A et al. (2019) Prospective detection of Brown H, Hammett T, Cantarini M and Barrett JC mutations in cerebrospinal fluid, pleural effusion, and (2015) EGFR mutation detection in ctDNA from ascites of advanced cancer patients to guide treatment NSCLC patient plasma: a cross-platform comparison of decisions. Mol Oncol 13, 2633–2645. leading technologies to support the clinical development 32 Ptashkin RN, Mandelker DL, Coombs CC, Bolton K, of AZD9291. Lung Cancer 90, 509–515. Yelskaya Z, Hyman DM, Solit DB, Baselga J, Arcila 23 Papadimitrakopoulou VA, Han JY, Ahn MJ, ME, Ladanyi M et al. (2018) Prevalence of clonal Ramalingam SS, Delmonte A, Hsia TC, Laskin J, hematopoiesis mutations in tumor-only clinical genomic Kim S-W, He Y, Tsai C-M et al. (2020) Epidermal profiling of solid tumors. JAMA Oncol 4, 1589–1593. growth factor receptor mutation analysis in tissue and 33 Coombs CC, Gillis NK, Tan X, Berg JS, Ball M, plasma from the AURA3 trial: osimertinib versus Balasis ME, Montgomery ND, Bolton KL, Parker JS, platinum-pemetrexed for T790M mutation-positive Mesa TE et al. (2018) Identification of clonal advanced non–small cell lung cancer. Cancer 126, hematopoiesis mutations in solid tumor patients 373–380. undergoing unpaired next-generation sequencing assays. 24 O’Leary B, Hrebien S, Beaney M, Fribbens C, Garcia- Clin Cancer Res 24, 5918–5924. Murillas I, Jiang J, Li Y, Huang Bartlett C, Andre F, Loibl S et al. (2019) Comparison of beaming and droplet digital PCR for circulating tumor DNA Supporting information analysis. Clin Chem 65, 1405–1413. Additional supporting information may be found 25 Li BT, Janku F, Jung B, Hou C, Madwani K, Alden R, online in the Supporting Information section at the end Razavi P, Reis-Filho JS, Shen R, Isbell JM et al. (2019) of the article. Ultra-deep next-generation sequencing of plasma cell- Fig. S1. Flowchart of the procedure work. free DNA in patients with advanced lung cancers: Fig. S2. Boxplot showing differences in MAFs between results from the actionable genome consortium. Ann patients progressing at the brain level exclusively and Oncol 30, 597–603. patients with disease progression assessed at other 26 Provencio M, Perez-Barrios C, Barquin M, Calvo V, anatomical locations. Franco F, Sanchez E, Sanchez R, Marsden D, Fig. S3. Comparison of MAFs obtained by NGS- Cristobal Sanchez J, Martin Acosta P et al. (2019) based platforms. Next-generation sequencing for tumor mutation quantification using liquid biopsies. Clin Chem Lab 58, Fig. S4. Boxplot showing differences in AFs between 306–313. sensitizing and the T790M mutation according to each 27 Barquın M, Maximiano C, Perez-Barrios C, Sanchez- platform. Herrero E, Soriano M, Colmena M, Garcıa- Table S1. Input volume and elution volume according Espantale on M, Tejerina Gonzalez E, Gutierrez L, to extraction method. Sanchez Ruiz AC et al. (2019) Peritoneal washing is Table S2. Analytical sensitivities for each platform. an adequate source for somatic BRCA1/2 mutation Table S3. Total amount of cfDNA (ng) obtained per testing in ovarian malignancies. Pathol Res Pract 215, mL of plasma for each patient and according to each 392–394. methodology. Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 55 18780261, 2021, 1, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12832 by Readcube (Labtiva Inc.), Wiley Online Library on [03/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Liquid biopsy-based cross-platform comparison A. Romero et al. Table S4. Agreement between methods for the detec- each methodology and according to the type of muta- tion of deletions in exon 19, point mutations in exon tion and considering the gold standard tumor biopsy 21 and the T790M mutation. obtained at diagnosis. Table S5. T790M detection rate in patients with and Data S1. List of all patients included in the study without CNS metastases according to method. showing mutation detected and corresponding mutant Table S6. Sensitivity, Specificity positive predictive allele frequency according to methodology. value (PPV) and negative predictive value (NPV) of 56 Molecular Oncology 15 (2021) 43–56 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd.